Background Ultrasound (US)-guided fine needle aspiration cytology (FNAC) is an accurate, reliable, and simple method to identify a thyroid nodule as benign or malignant. However, non-diagnostic cytology results for thyroid nodules are a major limitation of US-guided FNAC. Purpose To investigate the incidence of thyroid cancer among cases with non-diagnostic results on FNAC and to provide suggestions for the management of thyroid nodules that are initially non-diagnostic by FNAC according to ultrasonographic findings. Material and Methods From July 2006 to December 2009, 10,317 thyroid nodules in 6684 consecutive patients underwent US-guided FNAC at our institute. Among these, 871 thyroid nodules (8.4%) were diagnosed as non-diagnostic on initial cytologic evaluation and 196 underwent a second or third FNAC. Twenty-seven thyroid nodules (18.9%) underwent surgery, while 116 thyroid nodules were cytologically confirmed as benign with no remarkable change on follow-up US were included. We retrospectively reviewed the US findings for a total of 143 thyroid nodules (123 benign nodules and 20 malignant nodules). The US features that we compared included composition, echogenicity, margin, calcifications, shape, and underlying echogenicity. Results In total, thyroid cancer was diagnosed in 20 nodules (14.0%). The size of the nodule was significantly associated with malignancy ( P < 0.05). Most of the sonographically probable benign nodules were found to be benign (97.6%). Suspicious nodules on US were thyroid cancer in 43.2% of cases. Marked hypoechogenicity, microlobulated or irregular margin, microcalcifications, and taller-than-wide shape were significant US findings that correlated with malignancy ( P < 0.05). The diagnostic performance of ultrasound for initially non-diagnostic thyroid nodules was as follows: sensitivity of 90.0%, specificity of 65.0%, positive predictive value of 29.5%, and negative predictive value of 97.6%. Conclusion In terms of management of thyroid nodules with non-diagnostic FNAC cytology, US evaluation is a feasible and useful method for predicting malignancy.